epsl-data-analysis

Data Analysis (epsl-data-analysis)

EPSL's methods-transparency culture is strictest exactly where the journal is strongest: isotope geochemistry and geochronology. Reviewers expect every headline number to arrive with its complete error budget — internal precision, external reproducibility, standard and blank corrections, and systematic terms like decay-constant uncertainty — and they expect model results to arrive with their resolution limits. Design choices live in epsl-study-design; deposition and supplements in epsl-reporting-and-reproducibility.

When to trigger

• Reducing isotope, geochronologic, elemental, or geophysical data to reportable results
• Deciding which uncertainty terms to propagate and how to quote them
• Computing weighted means, isochrons, concordia intercepts, or Bayesian age–depth models
• A reviewer asked for the error budget, the MSWD, or a resolution test

Reporting norms EPSL expects

1. State the full uncertainty ladder. Quote internal (within-run) precision, external reproducibility from repeat standards/replicates, and — when comparing across methods or to other studies — systematic terms (decay constants, tracer calibration, spectrometer bias). Say which ladder rung each quoted ±X σ includes, and whether σ means 1σ, 2σ, or 95% CI.
2. Anchor to community values. Name the reference materials with the values used, the normalization scheme (e.g., what δ is measured against), and the decay constants adopted, with citations — so the numbers remain comparable when calibrations shift.
3. Statistics that match geochemical data. Weighted means with MSWD reported and interpreted (MSWD ≫ 1 means scatter beyond analytical error — say why); isochrons with the regression model named; mixtures and outliers handled by a stated rule, not silent deletion.
4. Age models with honest priors. Bayesian age–depth or eruption-tempo models report priors, convergence, and sensitivity to them; deposit the model input/config.
5. Inversions come with diagnostics. Resolution/recovery tests, misfit and trade-off curves, and an explicit statement of what the data cannot see; never present a smoothed model as if it were data.
6. Blanks and detection. Report total procedural blanks and their variability; state how blank-sensitive results (small samples, young ages, low abundances) respond to the blank range.

Uncertainty-reporting table reviewers work down

Element	What to report	If omitted, the reviewer assumes
σ-level and type	1σ/2σ/95% CI; internal vs external	numbers are not comparable
Reference materials	measured vs accepted values, per session	accuracy unverified
Blanks	magnitude + variability + correction	small samples untrustworthy
Decay constants / tracer	which values, cited	ages not portable across studies
MSWD / goodness of fit	value + interpretation	scatter hidden in the mean
Model resolution	recovery tests, trade-offs	structure may be artifact

Worked micro-example (illustrative — a weighted-mean age that survives review)

Twelve single-zircon U-Pb analyses from one ash bed (illustrative numbers):

• Ten young analyses give a weighted mean of 66.021 ± 0.024 Ma (2σ, analytical only), MSWD = 1.3 — scatter consistent with analytical error, so a single population is defensible.
• Two older grains are excluded as inherited by a pre-stated criterion (resolvably older than the main cluster), and shown in the figure anyway.
• Reported as: "66.021 ± 0.024/0.031/0.075 Ma (2σ: analytical / +tracer / +decay constants), MSWD 1.3, n = 10/12" — the three-tier bracket lets a reader compare against Ar-Ar or astrochronologic ages without emailing the authors.

The habit that prevents most queries: every mean is accompanied by its MSWD, its n-of-N, and a plot showing the excluded analyses.

Referee-pushback patterns and the venue-specific fix

• "Uncertainties are quoted but not defined." → State σ-level, and split analytical vs systematic terms wherever the result is compared to external ages or data.
• "MSWD indicates overdispersion." → Do not just widen errors; identify the geological or analytical source of scatter and let the interpretation absorb it.
• "The anomaly is at the edge of resolution." → Show the recovery test at that node; soften or drop claims the test cannot support.

Anti-patterns

• A headline age or rate with a bare ± and no statement of what it includes
• Standards measured but never reported against accepted values
• Outlier analyses removed without a stated rule or a visible plot
• MSWD omitted, or quoted without interpretation
• Tomographic/model features discussed where resolution tests show smearing
• Comparing your ages to literature ages without harmonizing decay constants

Output format

【Headline number】value ± (σ-level; analytical/+systematic) + units + n
【Traceability】standards vs accepted values; blanks; constants cited
【Statistics】weighted mean/isochron/Bayesian model + MSWD/diagnostics
【Exclusions】rule stated + shown in figures? [Y/N]
【Model diagnostics】resolution/sensitivity reported? [Y/N or N/A]
【Next】epsl-figures-and-tables

Supplementary resources

• ../../resources/external_tools.md — IsoplotR, ET_Redux, iolite, ObsPy, inversion codes
• ../../resources/official-source-map.md — methods-transparency and data expectations